Table 25 Macro-averaged results for dependency parsers on the CRAFT folds and dev set compared to untrained results on dev set; labeled attachment score (LAS), unlabeled attachment score (UAS), labeled accuracy score (LS)
Parser | Fold 0 | Fold 1 | Fold 2 | Fold 3 | Fold 4 | Training Average | Dev Set | Dev – WSJ model |
|---|---|---|---|---|---|---|---|---|
MaltParser | ||||||||
LAS | 88.45 | 88.70 | 89.62 | 89.12 | 88.85 | 88.97 | 88.93 | 72.40 |
UAS | 90.33 | 90.63 | 91.50 | 90.94 | 90.51 | 90.80 | 90.72 | 75.90 |
LS | 93.43 | 93.78 | 94.23 | 94.16 | 93.93 | 93.92 | 94.03 | 82.73 |
MSTParser | ||||||||
LAS | 88.30 | 88.85 | 89.58 | 89.12 | 88.90 | 88.98 | 89.36 | 75.99 |
UAS | 90.37 | 90.83 | 91.50 | 91.04 | 90.82 | 90.93 | 91.31 | 79.42 |
LS | 93.32 | 94.06 | 94.37 | 94.25 | 93.98 | 94.03 | 94.52 | 85.73 |
ClearParser | ||||||||
LAS | 89.09 | 89.43 | 90.33 | 89.86 | 89.59 | 89.68 | 90.09 | 74.56 |
UAS | 90.66 | 91.09 | 91.81 | 91.42 | 91.08 | 91.23 | 91.63 | 77.78 |
LS | 93.89 | 94.37 | 94.88 | 94.65 | 94.57 | 94.50 | 94.99 | 85.17 |